[Bug middle-end/123524] [16 Regression] 6% performance regression in gcc-16 compared to gcc-15 when compiling an interpreter

[pinskia at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123524

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(insn 82946 82944 82945 (set (reg:SI 99282)
        (const_int 131074 [0x20002])) "ipret-1.inc":1806:3 discrim 1 -1
     (nil))

(insn 82945 82946 82947 (set (reg:V4SI 99281)
        (vec_duplicate:V4SI (reg:SI 99282))) "ipret-1.inc":1806:3 discrim 1
9355 {*vec_dupv4si}
     (nil))

(insn 82947 82945 82948 (set (reg:V8HI 99280)
        (subreg:V8HI (reg:V4SI 99281) 0)) "ipret-1.inc":1806:3 discrim 1 -1
     (nil))

(insn 82948 82947 82949 (set (reg:V8HI 99279)
        (reg:V8HI 99280)) "ipret-1.inc":1806:3 discrim 1 -1
     (expr_list:REG_EQUAL (const_vector:V8HI [
                (const_int 2 [0x2]) repeated x8
            ])
        (nil)))

Wihch comes from:
  vect__63576.15000_96616 = vect__32948.14998_96624 & { 2, 2, 2, 2, 2, 2, 2, 2
};
  vect__63576.15000_96617 = vect__32948.14999_96615 & { 2, 2, 2, 2, 2, 2, 2, 2
};

So it is spilling to the stack.